Seventeen years since rimonabant's downfall: reassessing its suicidality risk profile
Abstract
Targeting the cannabinoid type 1 receptor (CB1) is a clinically validated antiobesity therapeutic approach. The only such drug approved, rimonabant, was launched in 2006 in Europe but subsequently rejected by the US Food and Drug Administration (FDA) in 2007. The FDA cited the increased risk of suicidality in its opposition to rimonabant's approval, leading to the drug's eventual worldwide withdrawal and the abandonment of this class of therapeutics. Seventeen years later, a new class of CB1-targeting drugs is emerging, but the impact of the 2007 FDA decision remains a formidable obstacle to its clinical development. We revisit the suicidality data presented by the FDA in light of the evolution of suicidality assessment and cross-reference this with the data in the subsequently published clinical trials. We conclude that the publicly available data do not support the FDA's conclusion that the use of rimonabant was associated with an increase in the risk of suicidality.
Study Importance
What is already known?
- Experimental drugs targeting the cannabinoid type 1 receptor (CB1) constituted a promising class of antiobesity medications in the early 2000s. The most advanced of these, rimonabant, was rejected by the US Food and Drug Administration (FDA) in 2007 over potential links to suicidality, leading to the abandonment of the entire class.
What does this review add?
- We revisit the suicidality data presented by the FDA in light of the evolution of suicidality assessment and cross-reference this with the data in the subsequently published clinical trials. We conclude that the publicly available data do not support the FDA's conclusion that the use of rimonabant was associated with an increase in the risk of suicidality.
- Reexamining the alleged link between rimonabant and suicidality is key to removing a perceived risk standing in the way of clinically developing this once-promising drug class.
INTRODUCTION
On June 13, 2007, the US Food and Drug Administration (FDA) Advisory Committee met to discuss the marketing authorization for Sanofi's (Paris, France) cannabinoid type 1 receptor (CB1) inverse agonist rimonabant (European Union brand name, Acomplia; United States proposed brand name, Zimutli) for the treatment of obesity [(1)]. The drug had been approved since 2006 in 37 countries, including the European Union [(2)]. In sharp contrast, the FDA advocated against the approval of rimonabant by citing the risk of psychiatric adverse events, with particular emphasis on suicidality. The Advisory Committee voted unanimously against the approval [(3)], and shortly thereafter the FDA formally rejected the marketing authorization request. In January 2009 [(4)], the European Medicines Authority (EMA) followed suit and reversed its 2006 approval of rimonabant, ordering the drug to be withdrawn. The entire drug class of CB1 inverse agonists was subsequently abandoned, with Pfizer, Bristol Myers Squibb, and Merck terminating their own late-stage clinical programs.
We believe that there is value in revisiting that decision and reexamining the data given the growing impact of obesity as a public health issue, as well as emerging evidence of the potential to combine CB1 receptor inverse agonism with incretin analogues [(5-7)]. Understanding the magnitude and context of the suicidality risk associated with rimonabant is essential to overcoming the reluctance by the pharmaceutical industry to develop a new generation of improved CB1 inverse agonists, an effort focused on peripherally restricted compounds with minimal brain exposure [(8-10)].
Herein, we analyze the data presented by the FDA to its Advisory Committee (including the briefing document, accompanying slides, and the meeting's transcript) [(11)] and cross-reference this with the relevant peer-reviewed literature pertaining to rimonabant clinical studies, much of it published after 2007. In this article, we focus solely on the 20-mg dose because this was the only dose submitted to the EMA and the FDA. We focus on reported completed suicides, suicide attempts, and suicidal ideation because the risk of these tragic phenomena was at the forefront of the FDA's objection and left a lasting stigma on the CB1-targeting drug class.
RANDOMIZED CLINICAL TRIALS VERSUS POST-MARKETING DATA
The FDA analyzed the following two sets of data: pooled data from randomized controlled trials (RCTs) and post-marketing data (real-world evidence [RWE]) from countries that had approved rimonabant for usage in patients with obesity. The cutoff for the data was December 18, 2006. The total pool of patients from RCTs corresponded to 14,832 patients from 13 pooled studies, of which 5998 were from Rimonabant in Obesity (RIO) studies. An additional 12,774 patients were in ongoing RCTs, and their data were still blinded. By the time the FDA Advisory Committee met on June 13, 2007, the numbers had increased (as more participants concluded ongoing studies), but the differences were relatively minor (e.g., 14,832 vs. 15,034 participants who had completed clinical studies). This paper will use the numbers cited in the FDA briefing book [(1)] unless otherwise stated. The FDA briefing book also reviewed 78,610 patients representing post-approval data from countries in which rimonabant had been launched. By June 13, 2007, the combined number of patients included in the briefing book had risen to 138,000, the vast majority of whom used rimonabant as an antiobesity drug either in the context of clinical studies or post-marketing usage.
SUICIDES
We reviewed the FDA briefing book [(1)], marking all references to completed suicides.
No completed suicides were listed in the pooled unblinded dataset of 13 rimonabant RCTs (n = 14,832) as of December 18, 2006 (Table 14 in the FDA briefing book).
Two completed suicides are referenced that occurred after the cutoff date of the submitted RCT data from Sanofi: one in RIO-North America in a patient who was taking rimonabant 5 mg and one in a then-ongoing study of coronary heart disease (STRADIVARIUS) in a patient who was taking rimonabant 20 mg.
- The RIO-North America study published by Van Gaal et al. [(12)] in 2008 clarifies the following: “One death was reported by the investigator and sponsor in the 5 mg rimonabant group as a completed suicide; however, after the independent blinded assessment of suicidality […] it was concluded that there was not enough information to classify this death as a ‘completed suicide.’” In the presentation on June 13, 2007, by Sanofi's Paul Chew, MD, the following further details were disclosed regarding this case: An adult male, aged 63 years, was found dead in front of his house with a gunshot wound. This was initially assumed to have been a suicide despite the record of his last visit, which noted “no sign of despondency, hopelessness or outward sign that the patient was suicidal.” Chew further disclosed that the patient did have “a past involvement in a federal witness protection program and was awaiting a criminal trial.”
-
The STRADIVARIUS publication by Nissen et al. [(13)] published in 2008 notes: “A single patient in the placebo group attempted suicide, and a single patient in the rimonabant group successfully completed suicide.” Chew's presentation discloses that this 36-year-old male individual became depressed 8 months into the rimonabant treatment, and that his depression worsened in the last 3 months of his life. The patient was facing financial problems and was overworked.
The two completed suicides mentioned by the FDA in the briefing book are thus reduced to one completed suicide and one possible homicide. That sole case of completed suicide occurred in the context of a non-obesity study in which the placebo and rimonabant cohorts each had a single suicide attempt. The placebo suicide attempt was (fortunately) unsuccessful, whereas the rimonabant one was (tragically) successful.
- The Comprehensive Rimonabant Evaluation Study of Cardiovascular Endpoints and Outcomes (CRESCENDO) study (cardiovascular events study): A 77-year-old male individual on rimonabant 20 mg
-
Post-marketing: A 33-year-old male individual on rimonabant 20 mg
- Chew highlighted the details of a completed suicide of a 77-year-old male individual: “About 10 months after study start, the patient became depressed. The patient had a history of depression dating back to his 40s. He discontinued rimonabant on his own. He visited a psychiatrist who prescribed SSRI [selective serotonin reuptake inhibitor] and committed suicide 1 week later. Psychiatrist evaluation revealed depression and loss of energy/interest [but] no associated suicidal ideation.”
- The post-marketing completed suicide was described as follows: “One post-market case of suicide reported (May 28, 2007) via secondhand information in a patient allegedly receiving rimonabant.” The patient was an adult male and had a body mass index (BMI) of less than 20 kg/m2. This unverified report marks the only on-record suicide in a post-marketing dataset of 138,044 patients, yielding a suicide rate of 0.0007%. It is unclear why an individual with a BMI of less than 20 was allegedly prescribed antiobesity therapy.
SUICIDALITY
Reliance on post hoc Columbia Classification Algorithm of Suicide Assessment measurement
The rimonabant clinical studies did not measure suicidality prospectively. The FDA's Division of Metabolism and Endocrinology Products (DMEP) therefore tasked Kelly Posner, PhD, and her colleagues at the Columbia University Department of Psychiatry to ipso facto analyze the clinical dataset and infer suicidality. Statistical analysis of their findings was conducted by the FDA's Division of Biometrics II. Posner and her team performed the analysis by applying the Columbia Classification Algorithm of Suicide Assessment (C-CASA) methodology [(14)], which classifies cases according to the categories of interest shown in Table 1.
| Classification | Assessment category |
|---|---|
| Suicidal | 1. Completed suicide |
| 2. Suicide attempt | |
| 3. Preparatory act toward imminent suicide | |
| 4. Suicidal ideation devoid of any preparatory action | |
| Indeterminate | 5. Self-injurious behavior intent unknown |
| 6. Not enough information: fatal | |
| 9. Not enough information: nonfatal | |
| Non-suicidal | 7. Self-injurious behavior without suicidal intent |
| 8. Other (accident, psychiatric, medical) |
- Abbreviation: C-CASA, Columbia Classification Algorithm of Suicide Assessment.
C-CASA was validated using post hoc analysis of clinical data from pediatric populations treated with antidepressants by reconstructing suicidal attempts using a word-capture algorithm from original case report forms in these clinical studies. Because this algorithm cannot capture suicidal ideation, C-CASA used independent, blinded expert raters to score this domain of the assessment. To quote from Posner et al.: “Expert review of cases was needed for inference of suicidal intent based on the details of behaviors and related clinical data, since many narratives lacked stated suicidal intent. Expertise in suicidality was determined by relevant experience and publications” [(14)].
Rimonabant had fewer suicide attempts than placebo in the pooled RCTs
Of the 14,832 pooled participants in the dataset, Posner and her team identified 74 cases of possible (4) or definite (70) suicidality. The majority were category 4, inferred retrospectively from the patient's record.
In Table 2, we replicate the FDA data on “definite suicidality” and add the percentage values with respect to total cohort population. No completed suicides were recorded in the pooled rimonabant clinical studies. More suicide attempts were recorded in the placebo group than in the 20-mg rimonabant group, whereas rimonabant 5 mg did not generate any suicide attempts. In contrast, the suicidal ideation category had an absolute increase of 0.12% in the 20-mg rimonabant cohort versus placebo. Paradoxically, the lowest number of suicidal ideations occurred in the 5-mg rimonabant cohort (threefold lower than placebo as a percentage of population).
| C-CASA classification | Placebo (n = 2909) | 5 mg (n = 5121) | 20 mg (n = 6802) |
|---|---|---|---|
| 1. Completed suicide | 0 | 0 | 0 |
| 2. Suicide attempt | 7 (0.24%) | 0 | 4 (0.06%) |
| 3. Preparatory act toward imminent suicide | 0 | 1 (0.02%) | 0 |
| 4. Suicidal ideation | 13 (0.44%) | 6 (0.12%) | 39 (0.56%) |
- Abbreviations: C-CASA, Columbia Classification Algorithm of Suicide Assessment; FDA, Food and Drug Administration.
No suicide attempts occurred in the obesity RCTs
Although the FDA provided the total suicidality numbers for the obesity studies, it did not provide the equivalent breakdown of these as it did for all pooled studies (including non-obesity). Fortunately, the data for the four largest obesity studies can be found in Van Gaal et al. [(12)] and account for 85% of the participants listed in Table 15 in the FDA briefing book [(1)] (three smaller and shorter phase 2 studies were not included).
The 11 suicide attempts (7 in placebo and 4 in 20-mg rimonabant) listed in Table 2 are absent in the Van Gaal et al. table. Therefore, these 11 events were attributable to non-obesity studies.
Finally, in an obesity study (RIO-Asia) [(15)] that was not included in the FDA analysis, a total of 643 participants were treated with either placebo or 20-mg rimonabant for 9 months. No incidents of deaths, completed suicides, suicide attempts, or suicidal ideation were recorded.
Non-obesity RCT suicidality data are skewed by a single smoking cessation outlier study
The FDA report provides additional granularity by listing the C-CASA suicidality data of the 13 individual studies (Table 3 is reproduced from the FDA briefing book Table 15). These data present an opportunity to explore the suicidality signal in non-obesity studies. These studies were included in the FDA statistical meta-analysis for rimonabant's obesity application despite the very divergent, and not necessarily comparable, patient populations that they represent.
| Study | Population | Placebo | Rimonabant | |
|---|---|---|---|---|
| 5 mg | 20 mg | |||
| ACT4855 | Alcoholicism | 3/127 (2.36) | 3/131 (2.29) | |
| METATRIAL | Schizophrenia | 7/98 (7.14) | 7/72 (9.72) | |
| EFC4474 | Smoking | 1/260 (0.38) | 1/256 (0.39) | 2/267 (0.75) |
| EFC4964 | Smoking | 1/261 (0.38) | 1/262 (0.38) | 0/261 |
| EFC5794 | Smoking | 1/268 (0.37) | 1/262 (0.38) | |
| EFC4796 | Smoking | 0/664 | 0/2016 | 12/3023 (0.40) |
| DRI3388 | Obesity | 0/73 | 0/67 | 1/69 (1.45) |
| EFC4733 | Obesity | 1/305 (0.33) | 0/603 | 6/599 (1.00) |
| EFC4735 | Obesity | 2/342 (0.58) | 2/345 (0.58) | 3/346 (0.87) |
| EFC4736 | Obesity | 0/348 | 0/358 | 2/339 (0.59) |
| EFC4743 | Obesity | 4/607 (0.66) | 4/1214 (0.33) | 7/1219 (0.57) |
| EFC5031 | Obesity | 0/80 | 1/76 (1.32) | |
| EFC5825 | Obesity | 0/140 | 1/138 (1.68) | |
- Abbreviation: FDA, Food and Drug Administration.
No measurable suicidality imbalances were seen in patients with alcoholism (study ACT4855 n = 258) or patients with schizophrenia (study METATRIAL n = 170), two populations that are prone to suicidality [(16, 17)]. Similarly, three of the four smoking studies (EFC4474, EFC5494, and EFC4964, total pooled 20-mg population of n = 790) yielded three cases of suicidality each for the placebo and the 20-mg rimonabant pooled cohorts, respectively. In contrast, study EFC4796 (STRATUS WORLDWIDE n = 3023) yielded 12 cases of suicidality in the 20-mg cohort and none in the placebo. Given the outsized impact of study EFC4796 on the suicidality statistical analysis, we wished to further examine the data from these smoking cessation studies. Data from smoking cessation studies were published by Robinson et al. [(18)] in 2018, 11 years after the withdrawal of rimonabant, but unfortunately exclude results of the outlier study EFC4796, which were apparently never published. Nevertheless, their insights are notable: There were no deaths reported in the three pooled studies, and the single suicide attempt occurred in the placebo cohort. The authors found no differences in the depression scores (a factor often important in retrospective C-CASA analysis) between placebo and rimonabant and highlighted the divergence in their findings from those of the FDA. Importantly, it was STRATUS WORLDWIDE that had an outsized impact on the FDA's determination of suicidality risk. In its analysis of this study, the FDA made the decision to treat the 5-mg arm as a control arm (combining it with the placebo cohort) despite Sanofi's position that this was an active-agent arm with measurable clinical efficacy and side effects (see study EFC4796 in Table 3).
Demographics of RCT suicidality data show European Union as an outlier potentially driven by differences in methodology
The FDA briefing book provides a breakdown of the 74 suicidality events by age, gender, time-to-event, BMI, and country (FDA briefing book Table 16). Within the 1-year pooled RIO studies, the US-only populations (65.5% of pooled placebo and 68.5% of pooled 20 mg, FDA briefing book Table 3) showed a similar absolute number of suicidal ideation events (six, five, and eight) with placebo and rimonabant (5 and 20 mg), respectively. Note that the 20-mg rimonabant group had 2.3 times more participants than placebo. Based on Van Gaal et al. [(12)], all the obesity suicidality cases listed in this table were suicidal ideations and not suicide attempts or completed suicides. However, in Europe, the suicidality data were skewed, with eight events in the 20-mg group, representing a near doubling of the event rate for 20-mg rimonabant compared with the United States, with the placebo notably lower (0%). These substantial differences in the reporting of events in Europe, ~30% of the enrolled RIO population, seem to account for the odds ratio differences highlighted in the FDA briefing book. No explanation is offered by the FDA for this discrepancy.
Christensen et al. [(19)] point out that, unlike the other three large RIO studies, RIO-Europe did not report changes in Hospital Admission Depression Score (HADS) from baseline. Recorded levels of depression and anxiety portions of the HADS were dramatically higher for both placebo and 20-mg rimonabant (5 and 20 mg combined), pointing to potential differences in assessment methodology. A comparison of RIO-Europe to RIO-North America (rimonabant values) yielded the following: 3.4 versus 0.1 (depression) and 5.6 versus 0.9 (anxiety). Placebo yielded similar differences: 2.7 versus 0.1 (depression) and 4.4 versus 0.2 (anxiety).
The results of the RIO-Europe study were covered in two publications [(20, 21)], both by Van Gaal et al. in 2005 (1-year data) and 2008 (2-year data), respectively. With no suicidality measured prospectively, we examined the depression scores from the prospective HADS as an imperfect surrogate. There were no meaningful changes in the depression score of HADS between placebo and 20-mg rimonabant (Table 7 in Van Gaal et al. [(20)]) that would seem to explain the retrospective finding by the FDA of zero suicidal ideations in placebo but eight such events in the 20-mg rimonabant cohort (Table 4). There is no mention of suicidality or completed suicides in the publication. The values for the second year readout [(21)] are even more striking, with a slightly higher rate of depressive events in placebo (6.2%) than rimonabant 20 mg (5.5%). Herein there are also no mentions of completed suicides or suicidal ideation.
| Placebo | Rimonabant | ||
|---|---|---|---|
| (n = 1895) | 5 mg (n = 2587) | 20 mg (n = 2786) | |
| US | 6 (0.48%) | 5 (0.28%) | 8 (0.42%) |
| Europe | 0 | 0 | 8 (0.98%) |
- Abbreviations: FDA, Food and Drug Administration; RIO, Rimonabant in Obesity.
The EMA also reviewed suicidal ideation when it approved rimonabant in 2006. No methodology is presented, but the EMA concluded that no differences were seen in suicidal ideation between placebo and rimonabant (see “The EMA”). This is in sharp contrast to the analysis of the same dataset conducted by the FDA.
Time-to-event was longer for the rimonabant cohorts than placebo
Time to C-CASA-determined suicidality events was longer for rimonabant 20 mg (mean = 98.7, median = 44.5) compared with placebo (mean = 63.6, median = 33). This is counterintuitive to what one would expect to see with an event-triggering agent. No explanation is offered to account for this. Whereas no case-by-case breakdown is presented for time-to-event, the minimum and maximum value for each is presented. Both placebo and rimonabant 20 mg have minimum values of “1,” meaning that at least one event occurred on the first day of dosing for one study participant in each cohort. It is hard to clinically reconcile how an event on the very first day of dosing (with either placebo or drug) could be related to the experimental agent. Conversely, the maximum time-to-event listed in the rimonabant cohorts is 610 days, which raises the same question.
Post-marketing data show markedly low number of suicidal ideation events and no completed suicides or suicide attempts
At the time of the FDA briefing book data cutoff, a total of 78,610 patients who were prescribed rimonabant as an approved drug were included in the analysis. No cases of suicide or attempted suicide were reported to the agency. A total of nine cases of suicidal ideation (six medically confirmed and three consumer reported) were on record, yielding a suicidal ideation rate of 0.01%. No information is provided on these nine cases, such as how the suicidal ideation manifested or was detected or whether they refer to nine individuals or fewer individuals but with multiple events. To put this in context, Graham et al. [(22)] report the suicidal ideation rate among young adults with obesity in the United States as 6.60%.
THE CRESCENDO STUDY
The CRESCENDO study was a global RCT study to explore the cardiovascular benefits of rimonabant [(23)]. Data from CRESCENDO are not specifically referenced in the FDA briefing book [(1)] because the study was only a third of the way through enrollment at the time the report was produced. As the largest rimonabant RCT study conducted with a total of 18,695 participants randomly assigned to receive either rimonabant 20 mg (n = 9381) or matching placebo (n = 9314), the study was prematurely terminated in July 2008. Of the 18,695 participants, only 341 (1.8%) had completed the study at the time of termination.
The authors noted, “[…] in this trial, regulatory authorities rather than the investigators, data and safety monitoring board, or sponsor, dictated its early discontinuation. In fact, in one country considering whether to discontinue all clinical research with rimonabant, the chair of this trial's data monitoring committee unsuccessfully argued the case for continuation of the trial” [(23)].
Topol et al. [(23)] provided minimal safety data from the study. They identified one completed suicide in the placebo cohort and four completed suicides in the rimonabant cohort with a corresponding five suicide attempts versus nine suicide attempts, respectively. No further information is provided on the circumstances of these tragic events or any details as to the deceased. It is impossible to assess whether these were indeed suicides let alone any causality to rimonabant, nor do we know whether they were clustered in certain centers or countries. It is unknown what the final suicidality data would have been had the remaining 98.2% of study participants been allowed to complete it and whether this imbalance reflects a real phenomenon or would have corrected itself over time.
It is hard to reconcile how the original dataset from the December 2006 FDA submission covering a total of over 107,000 patients yielded no completed suicides, but a single cardiovascular study with 9369 rimonabant-randomized participants (of whom only 162 completed the study) yielded five completed suicides.
It is noteworthy that the baseline rate of depression mood disorders in the CRESCENDO study was 15% for placebo and 15.4% for rimonabant 20 mg. This rate seems markedly higher than the rate in the general population, which the World Health Organization (WHO) estimates at around 5% (https://www.who.int/news-room/fact-sheets/detail/depression). This is even more surprising given that the clinical study had the following exclusion criteria: “Presence of any severe medical or psychological condition that, in the opinion of the investigator, would compromise the patient's safe participation, including uncontrolled serious psychiatric illness” [(23)]. No explanation is offered by either the study authors or the EMA to this markedly elevated baseline depression rate.
The EMA
Although the focus of this paper is on the FDA and its impactful decision in June 2007, it is worthwhile to look at the experience of the EMA with rimonabant. Following its earlier than expected approval in June 2006 [(24)] and the subsequent rejection by the FDA in June 2007 [(25)], the EMA conducted several reviews that eventually led to its withdrawal of marketing authorization for rimonabant in January 2009 [(4)].
These reviews provide a source of information as to suicidality as assessed by the EMA. In October of 2008 [(26)], the EMA provided an updated assessment of rimonabant. The section dealing with serious adverse events and deaths reported a total of five deaths (three in the 20-mg cohorts, one in the 5-mg cohort, and one in the placebo cohort). None was a suicide, and all were determined to be unrelated to rimonabant.
A marked change is seen in the January 2009 [(4)] assessment that was the basis for the drug's marketing authorization withdrawal. The EMA states that “in the finalized clinical studies, the incidence of suicidal ideation was similar between placebo groups and rimonabant with only one completed suicide.” That single suicide is presumably the one from the STRADIVARIUS study mentioned earlier. Paradoxically, the EMA then follows with the following statement: “However, in ongoing studies, six additional cases of completed suicide have been reported out of which five cases were treated with rimonabant. This is a serious safety concern.” These five completed suicide cases are presumably from the then-ongoing CRESCENDO cardiovascular study (although that study, published in 2010 by Topol et al., referred to just four) [(23)]. The EMA, similar to Topol et al., does not provide any information on these suicides.
Other CB1 inverse agonists
Although rimonabant was the only CB1 inverse agonist submitted for FDA review, several others were in clinical development at the time, including two in phase 3 studies: taranabant (Merck, NCT00131404 [high-dose study, n = 2502 patients] and NCT00420589 [low-dose study, n = 624]) [(27, 28)] and otenabant (Pfizer, NCT00483171, NCT00391196, NCT00396448, and NCT00375401; n = 5463 total patients). In all of these studies, there were no completed suicides reported. A single suicide attempt was recorded with taranabant in a patient with a history of suicidality. Interestingly, an otenabant phase 3 study (NCT00483171) used a revised prospective instrument, the Columbia Suicide Severity Rating Scale (C-SSRS) [(29)], to measure and monitor suicidality and suicidal ideation. The study was stopped early, and no patients completed the protocol. An end-of-study report [(30)] indicated no completed suicides or suicide attempts occurred, and suicidal ideation was limited to 5 patients overall (out of 498 total treatment patients), but no formal statistical analysis was provided. Overall, whereas the datasets are more limited than rimonabant and were never independently analyzed by the FDA, there appears to be no obvious signal with respect to completed suicide, suicide attempts, or suicidal ideation for these agents.
INCRETIN ANALOGUES AND SUICIDALITY
- Liraglutide [(31)]: Suicidality events were prospectively measured by the Patient Health Questionnaire (PHQ9) [(32)] across pooled phase 2 and 3 studies. Placebo yielded 2 out of 1843 participants (0.1%) versus liraglutide's 9 out of 3291 participants (0.3%). This included eight cases of suicidal ideation on liraglutide (vs. none in placebo) and one suicide attempt each for the placebo and liraglutide cohorts. In contrast, prospective C-SSRS analysis showed no difference (1% for placebo and liraglutide). The prescribing information for Saxenda (liraglutide) Section 5.8 specifically states, “Patients treated with SAXENDA should be monitored for the emergence or worsening of depression, suicidal thoughts or behavior, and/or any unusual changes in mood or behavior” (https://www.accessdata.fda.gov/drugsatfda_docs/label/2022/206321s015lbl.pdf).
- Semaglutide [(33)]: Suicidal ideation in pooled phase 3 studies were 2 out of 1529 for placebo (0.1%) and 5 out of 2650 (0.2%) for semaglutide. C-SSRS values were 0.6% for placebo and 0.4% for semaglutide, with one recorded suicide attempt for semaglutide. The conclusion by the FDA reviewer referencing the suicidality data was as follows: “There was no evidence to suggest an increased risk of psychiatric adverse events or suicidal behavior or ideation with semaglutide in the weight management program.”
- Tirzepatide [(34)]: In pooled phase 2 and 3 studies, adverse events of depression or suicidal ideation were reported in 19 patients (adjusted, 0.7%) in the comparator arms and 41 (adjusted, 0.8%) patients treated with tirzepatide. Suicide attempts were 1 for placebo (out of 2354) and 4 for tirzepatide (out of 5415). There was one instance of successfully completed suicide in the tirzepatide group. The conclusion from the FDA reviewer was as follows: “Generally, major imbalances in AEs [adverse events] of depression and/or suicidal ideation between the tirzepatide and comparator arms were not obvious. […] I do not feel that these AEs warrant inclusion in proposed tirzepatide labeling based on the data submitted to this NDA (new drug application)” [(34)].
It is worth reiterating that, in the entire RIO dataset for rimonabant, there were no suicide attempts and no completed suicides, and that the US sites (accounting for two-thirds of the participants) showed no difference in suicidal ideation as measured by C-CASA, nor did the EMA detect any imbalances in the incidence of suicidal ideation across the submitted clinical data that it reviewed [(4)].
| Drug name: patient numbers | Completed suicides | Suicide attempts | Suicidal ideation (%) | Methodology |
|---|---|---|---|---|
| Placebo: 1843 | 0 | 1 | 0 | PHQ9 |
| Liraglutide: 3291 | 0 | 1 | 8 (0.24%) | |
| Placebo: 1539 | 0 | 0 | 2 (0.12%) | PHQ9 |
| Semaglutide: 2650 | 0 | 1 | 5 (0.18%) | |
| Placebo: 2354 | 0 | 1 | 19 (0.81%) | PHQ9 |
| Tirzepatide: 5415 | 1 | 4 | 41 (0.75%) | |
| Placebo: 1602 | 0 | 0 | 8 (0.50%) | C-CASA |
| Rimonabant 20 mg (RIO): 2503 | 0 | 0 | 17 (0.67%) |
- Abbreviations: C-CASA, Columbia Classification Algorithm of Suicide Assessment; PHQ9, Patient Health Questionnaire 9; RIO, Rimonabant in Obesity.
CONCERN AROUND SUICIDAL IDEATION RETURNS TO HAUNT INCRETIN ANALOGUES IN THE EUROPEAN UNION
On July 10, 2023, the EMA stated that it was looking at adverse events noted by the Icelandic Medicines Agency, including two cases of suicidal thoughts linked to the drugs Saxenda and Ozempic. One additional case relating to thoughts of self-injury has been raised in connection with Saxenda [(35)]. On July 11, 2023, the EMA announced that it was expending this investigation to other glucagon-like peptide-1 (GLP-1) receptor agonists and was analyzing 150 reports of possible suicidality [(36)]. Within days, the British Medicines and Healthcare Products Regulatory Agency (MHRA) announced a similar such investigation encompassing tirzepatide as well [(37)]. The FDA Adverse Event Reporting System (FAERS) in the July through September 2023 quarterly review similarly stated that the FDA is evaluating the need for regulatory action [(38)]. In early 2024, the FDA released its preliminary communication finding that GLP-1 receptor agonists did not demonstrate a clear relationship between their use and suicidal thoughts or actions but that it “is continuing to look into this issue” [(39)].
DISCUSSION
The putative link between rimonabant and an increased risk of suicidality was the determining factor in the drug's rejection by the FDA and subsequent withdrawal worldwide. It is the most formidable obstacle to the advancement of novel CB1 inverse agonists. Understandably, pharmaceutical companies are reluctant to take on the risk of developing a drug that could lead to such a tragic outcome.
- No completed suicides were recorded in the rimonabant RCT study or post-marketing data packages up to the cutoff date of December 2006 (n = 107,000).
- Suicide attempts were more prevalent for placebo than 20-mg rimonabant in the totality of pooled studies.
- No suicide attempts were recorded in the obesity (i.e., RIO) clinical studies.
- In US participants, retrospectively determined suicidal ideation was more common in the placebo group compared with 20-mg rimonabant. This is reversed in a single RIO-Europe study, and an outlier smoking cessation study (also in Europe) contributed further to this perception. In contrast, the EMA did not report any imbalance in suicidal ideation for the same clinical dataset.
- Time to suicidality events was shorter for placebo than for rimonabant in the pooled clinical data.
- None of the obesity studies analyzed by the FDA biostatistics division generated a statistical significance regardless of the analysis undertaken.
The use of retrospective C-CASA analysis, often years after the fact, raises concerns around the risk of “false positives.” C-CASA was developed as a tool to measure suicidality in children and teenagers [(14)]. It was not validated for obesity studies and was replaced by the FDA in 2012 with the prospectively measured C-SSRS [(40)]. Notably, the authors of the 2006 paper by the FDA on the use of this identical statistical analysis technique retrospectively in antidepressants in pediatric populations raised serious concerns around their own methodology [(41)]. Posner also voiced concerns regarding C-CASA, referencing rimonabant specifically, in 2008 [(42)] and 2009 [(43)]. Additional reservations were subsequently raised in 2010 by an internal review conducted by Mitchell Mathis, MD, Deputy Director, Division of Psychiatry Products, at the FDA [(44)], concluding that the post hoc use of meta-analyses of suicidality data (as was the case for rimonabant) was “…not likely to be direct source of evidence pertinent to the risk of suicide.”
In a contemporaneous FDA review, the smoking cessation drug varenicline (Chantyx) was also subjected to retrospective C-CASA analysis, leading to the FDA finding a risk of suicidality and the inclusion of a black box on the label in 2009. A phase 4 RCT study now using the prospective C-SSRS instrument found no such risk, leading to the FDA to remove the black box warning in 2016 [(45)]. More recently, Klein et al. [(46)] reanalyzed the RCT data for five antiseizure medications that were subjected previously to C-CASA post hoc analysis by the FDA and were found to be linked to suicidality. The reanalysis concluded that not one of these drugs was linked to suicidality.
In stark contrast, data presented by the FDA for the approval of liraglutide [(31)], semaglutide [(33)], and tirzepatide [(34)], albeit with different methodologies (prospective PHQ9 and C-SSRS), yielded similar values for suicidal ideation and more suicide attempts than were generated by rimonabant; however, the conclusion of the reviewers in all three cases was that the drugs merited approval. Viewed in that context, the risk for rimonabant may be equal to or less than what is seen with incretin analogues, but there appears to be enough evidence, including in the original FDA report, to question the link among rimonabant and suicidal ideation, self-harm events, and suicidal attempts/completed suicides. This also bodes well for the risk/reward dynamic of the potential of combining these two orthogonal mechanisms of actions into a single course of therapy. The additional weight loss that could be gained, and is supported by preclinical data, could be impactful but without adding significant risk related to the respective therapeutic agents.
We find it difficult to reconcile the data presented by the FDA (and the corresponding publications) with the FDA's conclusions. Factoring in the potentially flawed methodology (since abandoned by FDA) [(47)] and in the context of other antiobesity drugs that have since been approved, the hypothesis that rimonabant was associated with suicidality seems null. Unfortunately, the outcome of this rejection affected not only rimonabant but an entire drug class. Basic research into the role of CB1 in metabolism and endocrine pathways continues to this day with multiple publications focused on novel CB1 inverse agonists, all aiming to avoid the perceived pitfalls of rimonabant and its peers [(8-10, 48-51)]. However, there is reluctance from the industry to invest in this mechanism of action for understandable reasons given the FDA's rejection. Worse yet, the FDA provided no roadmap for a way forward; in effect, it highlighted a risk based on either no events (completed suicides), fewer events than seen in placebo (suicide attempts), or exceedingly rare events retrospectively determined by a now abandoned methodology (suicidal ideation seen in RIO-Europe) without delineating a threshold for acceptable safety. The evolution of the FDA's approach and the introduction of improved prospective instruments for assessing suicide risk point to an opportunity to more accurately assess the risk–benefit profile of this important drug class.
CONFLICT OF INTEREST STATEMENT
All authors listed are current or past employees of Corbus Pharmaceuticals Holdings, Inc.
